Witness statement transcription and chain of custody

could see it, and can you prove deletion when you say you deleted it?

Authentication at trial: what actually gets admitted

Three things might end up in front of a jury, and they're authenticated differently.

The recording itself

This is the primary exhibit. The witness — or the person who made the recording — testifies that they captured the statement, that the audio is unaltered, and that what's about to be played is what was said. Your custody log is the backbone of that testimony.

If the witness later recants or claims the statement was edited, your hash chain is part of the rebuttal. "Here is the SHA-256 hash computed at 14:32 on the day of the interview. Here is the hash of the file we're playing today. They match." That only works if the first hash was computed and logged before any processing.

The transcript as a demonstrative

Most courts allow a transcript to be handed to the jury as a reading aid while the audio plays, with a limiting instruction that the audio — not the text — is the evidence. To use an AI transcript this way, you typically need:

• A stipulation with opposing counsel that the transcript is substantially accurate, or
• A foundation witness (often a paralegal or the attorney) testifying that they compared the transcript line by line against the audio and corrected discrepancies.

The second path is where AI saves time and creates risk. The time savings are real — a 90-minute interview that would take 6+ hours to type can often be reviewed and corrected in 90 to 120 minutes. The risk: if you don't actually review every line, an ASR error becomes your error when you certify or offer the transcript.

A certified transcript as substantive evidence

If you need the transcript itself to be substantive evidence — for example, in a sworn statement context, or where the recording is unavailable — you generally need a human transcriptionist's certification or a court reporter. AI alone doesn't produce a certifiable transcript. We don't pretend it does, and neither should Otter, Rev's AI tier, Trint, or any other automated service. Rev's human transcription tier is a separate product and can produce a certified transcript if ordered; if you need that, use it.

A workflow for a small firm without an evidence-management system

If you're a 5-attorney shop without Relativity or Nuix, here's a workable pattern. Lawyers using AI transcription typically converge on something like this.

1. Capture on a single dedicated device per matter — a Zoom H5, an iPhone with Voice Memos, whatever — and never use that device for anything else during the matter. Record separate channels when you can: audio with each speaker on a separate channel is much easier to attribute by speaker, while mono diarization tools become less reliable as speakers overlap or the speaker count rises.
2. At the start of the recording, state the date, time, timezone, matter, participants, and recording method on tape. Have each speaker identify themselves. This is the cheapest authentication step you will ever take.
3. Immediately after the interview, transfer the file to a secure evidence folder or repository (your firm's encrypted cloud or evidence storage, not personal Dropbox, and not an unmanaged auto-synced directory). Compute the SHA-256 hash. Mark the original read-only. Record both in a custody log spreadsheet kept with the case file.
4. Upload a copy — never the original — to your transcription service. Log the upload timestamp, the model and version or vendor-exposed identifier, and the resulting transcript filename.
5. Save the raw machine output as v1_ai_raw and don't overwrite it. Review against the audio. Save the corrected transcript as v2_reviewed_[initials]_[12May2026] with its own hash.
6. Log every subsequent access. "12 May 2026, 10:14 — opened by paralegal J. Ortiz for redaction of personal identifiers." Log every production to opposing counsel, experts, or the court reporter, with date and confidentiality terms.
7. Before any deposition or motion citing the transcript, re-verify the hash of the original against your initial log entry.

Export final transcripts in formats that outlive your subscription — TXT, SRT, VTT — not proprietary containers a future vendor change will lock you out of. This is roughly the same discipline a forensic examiner applies to a hard drive image. It's tedious. It's also the difference between a transcript that survives a motion to exclude and one that doesn't.

Deposition prep is where AI earns its keep

For deposition preparation, the rules relax, but not to zero. A witness statement transcript used to prep a deponent, to draft a chronology, or to identify inconsistencies across 12 prior statements may never have to be admitted as evidence. You're usually using it as work product, though materials shown to a witness can raise discoverability or memory-refreshing issues under rules like FRE 612 and state analogs.

This is where AI transcription does its highest-value work for litigators: ingesting hours of prior statements, recorded calls, voicemails, and 911 audio into searchable text, fast. Search "blue truck" across the corpus and find every mention in seconds. Use the transcript as an index into the audio — find the timestamp, then listen before you quote. Treat the first AI draft as a map, not a quote bank.

Useful prep outputs include timeline bullets with timestamps, topic indexes, candidate impeachment excerpts, name and entity lists, and flagged inaudible sections needing follow-up. Keep prep transcripts clearly labeled and separate from any transcript you intend to offer or stipulate to at trial — different folder, different naming convention, different access controls if you can manage it.

What next

• Build a one-page custody log template before your next interview — file name, hash, capture device, capture time, every transfer, every access. A Google Sheet is fine; the audit trail is what matters, not the tool.
• Try a Free plan upload with a non-privileged practice recording to see what the raw transcript looks like before you commit to a workflow.
• If you handle witness audio over the phone, run a short test on 8 kHz audio so you know what the error rate feels like in your subject matter — medical malpractice vocabulary fails differently than commercial-contract vocabulary.
• If you need a certified deposition record or a forensic custody affidavit, use a court reporter, forensic examiner, or evidence-management provider. We can transcribe the working copy; we should not be your evidence custodian.

The transcript doesn't have to be the certified record to be useful. It has to be honest about what it is — a fast, searchable, machine-derived view of the audio — and it has to live inside a custody trail that survives a hostile cross.